Incidence and remission of asthma and respiratory symptoms in adults
The Hordaland County Cohort Study
Tomas Mikal Lind Eagan, MD
Department of Thoracic Medicine, Haukeland University Hospital,
University of Bergen, Norway 2004
ACKNOWLEDGEMENTS 4
LIST OF PAPERS 6
TERMS AND ABBREVIATIONS 7
INTRODUCTION 9
Prevalence 10
Incidence 12
Remission 17
Risk factors 17
AIMS 19
MATERIALS AND METHODS 20
Study design 20
Data collection 21
Processing the data file 22
Inconsistencies 22
Typing errors 23
Missing values 24
Variables 24
Statistical analyses 25
SYNOPSIS OF PAPERS 27
Paper I 27
Paper II 28
Paper III 29
Paper IV 30
Paper V 31
DISCUSSION 33
Methodological considerations 33
Study design 33
Wording of the questions 34
Statistical methods 35
Internal validity 38
Selection bias 38
Confounding 39
Information bias 40
External validity 45
Discussion of main results 48
Measures of incidence 48
Incidence rates - asthma 49
Incidence rates – respiratory symptoms 51
Age trends 52
Risk factors for the incidence of respiratory symptoms and asthma 53
Smoking 53
Biological mechanisms 56
Inflammatory changes in subjects with asthma 56
Inflammatory changes in smokers 57
Smoking related to asthma severity 58
Occupational exposure 59
Socioeconomic status 62
Perspectives 66
CONCLUSIONS 69
REFERENCES 71
APPENDICES 79
Appendix A - The baseline questionnaire and two reminder letters 79 Appendix B - The follow-up questionnaire and the accompanying letters 86 Appendix C - Analyses of incidence of wheeze using the question on wheeze within the last 12
months at follow-up, compared to wheeze ever 101
Appendix D - Example of the effect of a hypothetical reporting error on the estimates of incidence
and remission 102
Appendix E - The wording of the questions, or case definitions, used to determine self-reported asthma and respiratory symptoms in previous studies on asthma and symptom incidence 105
PAPERS 110
ACKNOWLEDGEMENTS
Large studies of the general population, followed for many years, obviously require a great effort. When the planning of the Hordaland County Study started, I was still in primary school. The Hordaland County Study is the brainchild of my second supervisor, professor Amund Gulsvik. Through relentless effort, and with sincere interest, he is largely responsible for the creation of the epidemiologic research environment at the Department of Thoracic Medicine. Not many weeks had passed after I started at the department, before he pulled me aside and told me he trusted I would want to seek answers only available through real science. We all know Amund as a man who follows up on our work closely, with great focus, support and eagerness. For the last part of the time I have been working on this project, he has remained almost uncharacteristically in the background, although always available for help and comments. I take that as a compliment of trust.
Of course, most of my training has come from my first supervisor, professor Per Bakke.
Per is widely known among research fellows, far beyond the Department of Thoracic Medicine, as one of the best supervisors around. I know this first-hand. A popular man, Per is always short of time, but never fails to give me the time I need, no matter how inconvenient. A mild man, he never presses me when he knows better; which I think is a strategy of allowing me to come to my senses by myself. And, Per is a very skilled man, in treating people so they feel valuable, and in treating epidemiology so it makes sense.
My third supervisor is associate professor Geir Egil Eide. A statistician with particular knowledge of attributable fractions, he has provided the gravity needed to focus our analyses, both before and during. Geir has always been willing to teach me the how’s of analyzing, although it sometimes took more than one effort. And Geir is perhaps my best critic when writing in English; once he has approved a manuscript, I feel certain few errors remain.
I feel fortunate, as my three supervisors have complemented each other, helped me grow, and stood up for me. Thank you.
With data collection starting almost 20 years ago, I fear not paying tribute to all those participating. In alphabetical order, Eli Henricksen, Borghild Hovland, Peer Lilleng, Bjørg Meidell, Steinar Nielsen, Ernst Omenaas, Olav Overå, Sølvi Sletten, Lene
Svendsen, and Ida Welle have all had important or vital roles in the data collection of the baseline survey or follow-up survey or both. The Department of Thoracic Medicine relies upon the combined effort of all those regularly involved, and then some. On behalf of the department I wish to express my thanks to all those who help create the positive research environment at the department, by readily helping out, even when not directly involved.
I also want to thank the Department of Thoracic Medicine itself, for enabling my work by means of paying my salary, while I have been working 50% clinically, and 50% on this dissertation.
Friends and colleagues are invaluable in sharing knowledge, frustrations and sometimes beer. I think you know who you are, but Tor Arne Strand, Jan Christian Brøgger, and Jon Andrew Hardie deserve special mention.
At times I have been absent-minded and remote, when at home. The love, patience and support I have received from Mikal, Erik and Christine is what makes everything worthwhile.
This study (baseline and follow-up survey) has been financed foremost by grants from the Norwegian Research Council, the Norwegian Asthma and Allergy Association, and the Royal Norwegian Council for Scientific and Industrial Research, with additional grants from GlaxoSmithKline, and AstraZeneca.
I have worked at the Department of Thoracic Medicine since 1998 in a 50% clinical and 50% research position, as part of the Department’s ongoing support of active research. I have received a research grant from GlaxoSmithKline enabling equipment of
computation. To present the results of my research at the American Thoracic Society (ATS), and European Respiratory Society (ERS) annual meetings, I have received support from GlaxoSmithKline, AstraZeneca, and the Norwegian Thoracic Society.
LIST OF PAPERS
Paper I
Eagan TML, Eide GE, Gulsvik A, Bakke PS. 2002.
Nonresponse in a community cohort study. Predictors and consequences for exposure- disease associations.
Journal of Clinical Epidemiology; 55(8): 775-81.
Paper II
Eagan TML, Bakke PS, Eide GE, Gulsvik A. 2002.
Incidence of asthma and respiratory symptoms by sex, age and smoking in a community study.
European Respiratory Journal; 19(4): 599-605.
Paper III
Eagan TML, Gulsvik A, Eide GE, Bakke PS. 2002.
Occupational airborne exposure and the incidence of respiratory symptoms and asthma.
American Journal of Respiratory and Critical Care Medicine; 166(7): 933-8.
Paper IV
Eagan TML, Gulsvik A, Eide GE, Bakke PS. 2004.
Remission of respiratory symptoms by smoking and occupational exposure in a cohort study.
European Respiratory Journal; 23(4): 589-94.
Paper V
Eagan TML, Gulsvik A, Eide GE, Bakke PS. 2004.
The effect of educational level on the incidence of asthma and respiratory symptoms.
Respiratory Medicine; 98(8): 730-6.
TERMS AND ABBREVIATIONS
Terms:
Prevalence - The proportion of a population that has disease (or an event like a symptom) at a specific point in time. The time can be immediate, a point prevalence, or over a longer time period, like for instance lifetime prevalence.
Incidence can be presented as Incidence rates, or Cumulative Incidence:
Incidence Rate - The occurrence of new disease per unit of person-time.
Cumulative Incidence - The proportion of people who develop new disease during a specified period of time (a.k.a. incidence proportion).
Inversely,
Remission rate - The occurrence of a return to a non-disease state per unit of person-time.
Cumulative remission - The proportion of people who return to a non-disease state during a specified period of time.
Prospective study - A study in which the information about exposure(s) was gathered before the studied outcome occurred.
Retrospective study - A study in which the information about exposure(s) was gathered after the studied outcome occurred.
Cross-sectional study - A study where all information is gathered at one point in time.
Prevalence study - A cross-sectional study conducted to estimate prevalence is often called a prevalence study.
Longitudinal study - A study where the information is gathered over a period of time.
Cohort study - A study where a defined group of exposed and unexposed subjects is followed for a set amount of (follow-up) time, and outcomes are recorded within and/or after the follow-up time.
Incidence study - A cohort study conducted to estimate incidence is often called an incidence study.
Odds - The ratio of the probability that some event will occur over the probability that the same event will not occur.
Odds ratio - The ratio of the odds in one group of subjects over the odds in another group of subjects.
Attributable fraction - The fraction of new cases that would have been prevented had the exposure in question not occurred.
Logistic regression - A mathematical modeling approach used to describe the effects of multiple explanatory variables on a dichotomous response variable, based on the logistic model.
Logistic model - a statistical model of an individual’s probability of developing an outcome as a function of the possible risk factorsX:
The linear term ß0+ß1X1+ß2X2+...+ßkXk is called the logit.
Abbreviations:
AF - Attributable Fraction BAL - Broncho-Alveolar Lavage
BHR - Bronchial Hyper-Responsiveness
BMI - Body Mass Index
CD4 - Cellular Determinant protein 4 (T helper cell) CDC - Centers for Disease Control
CI - Confidence Interval
COPD - Chronic Obstructive Pulmonary Disease EBC - Exhaled Breath Condensate
ECRHS - European Community Respiratory Health Survey FEV1 - Forced Expiratory Volume in 1 second
HMW - High Molecular Weight
HUNT - Helse Undersøkelsen i Nord Trøndelag IgE - Immunoglobulin, class E
IL-# - Interleukin #
IUATLD - International Union Against Tuberculosis and Lung Disease LMW - Low Molecular Weight
MeSH - Medical Subject Headings
OLIN - Obstructive Lung disease In Northern Sweden study
OR - Odds Ratio
SAS - Statistical Analyses Systems
SPSS - Statistical Product and Service Solutions Th1/Th2 - T helper cell class 1 and 2
ß0+ß1X1+ß2X2+...+ßkXk
e
ß0+ß1X1+ß2X2+...+ßkXk
1
+ eP(D=1| X1,X2,...,Xk) =
INTRODUCTION
In approximately the fifth century AC, the Greco-Roman physician Aurelianus Caelius phrased one of the earliest working descriptions of asthma (1):
“Asthma occurs oftener in men than in women, in middle age than in children or old men, and in the delicate rather than in the strong. More in winter than in summer and more at night than by day. In some it begins after disease, whereas in others it begins without obvious cause... the patient has a feeling of suffocation, heaviness and burning heat of the chest, and a feeling of spasm in the bowels. It begins with violent suffering, wheezing in the chest, and the voice is weak, the neck and face stretched and red, the expression anxious...there are tears...and the pulse is weak. Asthma is distinct from other diseases where there is difficult breathing...”
Central to our understanding of a disease is its distribution, as the first part of Aurelianus Caelius’ description so clearly shows. Apparently, the sex- and age- distribution of a disease has been a diagnostic help for many centuries. However, Caelius’ description also hints at seasonal variations of asthma, and a risk factor as an explanatory variable,
namely a prior disease. Presumably this could be an airways infection, an occurrence no less common in the times of the Roman Empire. Finally, this description underscores how the diagnosis of asthma is dependent on observation of its symptoms, symptoms not necessarily specific for asthma.
Epidemiology can be defined as: “the study of the distribution and determinants of health-related states or events in specified populations and the applications of this study to control of health problems” (2). Included in this definition is not only the description of the occurrence of disease, but also the preventive purpose of the investigation.
The prevalence and incidence of a disease are the most central measures of occurrence in epidemiology. The prevalence is a point-measure in time, and is more easily measured than incidence, which requires a time-period of observation. This is reflected in the medical literature, where there are many times more papers describing the prevalence of any given disease than its incidence.
The measurement of respiratory symptoms can be part of the measurement of asthma.
But measuring each symptom individually is attractive for other reasons. The perception of dyspnea or cough may not vary over time the same way as a diagnostic label, like asthma. Thus, some of the inherent bias one expects with self-reported asthma, may be avoided when asking specifically about a single symptom. The downside is the lack of specificity in the symptoms; for instance is dyspnea common in a wide range of diseases.
This can partly be overcome by examining combinations of symptoms, like attacks of dyspnea and wheezing.
When evaluating the course of a disease over time, knowing the normally occurring distribution and development of a symptom can be helpful. Also, the effect of a risk factor may not be the same on dyspnea as on cough, although both symptoms can be part of the same diagnosis.
The three most common symptoms in asthma are cough, dyspnea and wheezing. Further distinction is usually sought, like whether the cough is productive or degrees of dyspnea.
The most commonly examined respiratory symptoms in community health surveys are phlegm when coughing, dyspnea in varying degrees of exertion, attacks of dyspnea, and plain wheezing.
Prevalence
In Norway, four major studies from 1972 to 1998/99 have estimated the prevalence of asthma and respiratory symptoms in the adult general population (3-6). In the first phase of the Hordaland County Study on respiratory health in 1985, the adult prevalence of asthma was 3.5% with no difference between genders (7). The prevalence of phlegm cough and wheezing was higher in men than in women, 25%-17% for phlegm cough and 23%-19% for wheezing respectively (7). The prevalence of dyspnea grade 2 was higher in women, 13% versus 10% in men.
In the Nord-Trøndelag Health Survey (HUNT), the adult prevalence of asthma was 8.6%, and not significantly different in between genders (8). The prevalence of phlegm cough was roughly 8%, and daytime coughing approximately 15% (8).
In the most recent study, Brøgger et al examined the prevalence of attacks of dyspnea, wheezing, and asthma in Western Norway and Oslo, and compared them to the
prevalences found in the earliest study, by Gulsvik in Oslo in 1972 (9). The prevalence of both symptoms and asthma had increased since 1972. In 1998/99 the prevalence of asthma was 7.6% in men, and 10.7% in women. The prevalence of wheezing was roughly 25% for both genders, whereas the prevalence of attacks of breathlessness was higher in women, 19%, versus 14% in men (9).
The prevalence of asthma varies greatly between countries. This is unlikely to be due only to methodological and diagnostic differences (10). In the European Community Respiratory Health Survey (ECRHS) conducted between 1991/93, the asthma prevalence varied from 2% in Tartu, Estonia, to 12% in Melbourne, Australia, with a median
prevalence of 4.5%. The prevalence was highest in Australia, New Zealand, and the UK (11). In the Norwegian part of the ECRHS, with a study sample from Bergen, the asthma prevalence was found to be 4.3% (11). Further, wheeze within the last 12 months varied between 4% in Bombay, India to 32% in Dublin, Ireland, with a median of 20.7% (11).
In the last two decades, a great deal of attention has been given to the perceived rise in asthma prevalence, at least in the Western countries. This has been studied more
extensively in children. Although there is some evidence of changing diagnostic labeling (12), the vast number of studies point towards a real increase in asthma prevalence, also among adults (6).
In a stationary population, the relationship between the prevalence and the incidence of a disease can be described as (13):
Prevalence = Incidence * Duration of disease
If we define asthma as a chronic, or life-long disease, it follows that an increase in asthma prevalence should be due to an increase in asthma incidence. It should be noted that this is a simplification, and that the exact relationship between prevalence and incidence of asthma is not yet clearly defined (14).
Incidence
Before we can even approach the question of a possible increase in the incidence of asthma, we need to know what the adult incidence of asthma is. Unfortunately, there are few longitudinal studies from general populations, on the incidence of respiratory symptoms and asthma in adults.
While measuring the incidence of asthma and/or respiratory symptoms, three main designs have been applied: First, the cohort studies, which follow a cohort of subjects for a set period of time, and count the new events. Second, the cross-sectional studies, where subjects have been asked when in time they first were diagnosed or experienced
symptoms.1 Finally, there are register-based studies, using medical registers to obtain diagnoses in the population to which the register pertains.
An overview of the main studies on adult incidence of asthma and respiratory symptoms is given in table 1. Studies were identified through MEDLINE, with searches consisting of various combinations of the following keywords: respiratory symptoms, cough, dyspnea, wheezing, asthma, incidence, cohort, longitudinal, community, general population, and the MeSH terms: ‘Cohort Studies’, ‘Asthma’, ‘Signs and Symptoms, Respiratory’. Only English language papers were evaluated. From the bibliographies of retrieved papers, additional relevant papers were identified. All cohort studies on general populations were included (15-29), if calculations of incidence rates of asthma and/or respiratory symptoms were possible. Further, important cohort studies from selected populations, like the British National Child Development Study (30), the Finnish Twin Cohort Study (31, 32), and the Nurses Health Study (33) were included, as well as four large cross-sectional surveys (34-37), and one register-based survey (38).
The number of published articles presenting incidence rates is greater than the number of studies. For instance, the cohort from Tucson, Arizona had 10 follow-ups up until 1988.
1 These are sometimes referred to as retrospective cohort studies. I find that somewhat confusing, and the term cohort study is used in this dissertation to denote a prospective study design.
Three papers have published incidence rates from this study, one based on the first three follow-ups (15), one on baseline, the fifth and the eighth follow-up (16), and one on baseline and the eighth or ninth follow-up (17). If we consider only the methodology, and counts for instance the Tucson, Arizona cohort for one study, then there are 21 studies in table 1.
Twelve studies used a random selection of the source population (table 1), and for two studies an entire geographical area was defined as the source population (23, 38).
However, two studies did not give a clear account of the selection (18, 24). In eleven of the studies, information was collected by interview (15-18, 21, 23-26, 29, 30, 35). In none of these eleven studies have the numbers of interviewers involved been reported.
Six of the studies did not report the exact wording of the questions used to define asthma or respiratory symptoms (table 1) (17, 18, 21, 23-25).
Table 1. Studies estimating adult incidence of asthma and respiratory symptoms
DesignStudy Name Location Author, year Ref #
Years Study was conducted
Years on which incidence rates are based
Years of follow- up
Random sampling of source population
Age-span of study population
Self- employed questionnaire
/Interview Number of interviewers
Tucson,
Arizona, USADodge, 80 15 1972 - 1988 1972/73-
1976 3.5 no* all ages nurse-led
interview not reported
Tucson,
Arizona, USAKrzyzanowski,
90 16 1972 - 1988 1972/73 -
1983/84 11 no*† 19-70 nurse-led
interview not reported
Tucson,
Arizona, USAKrzyzanowski,
92 17 1972 - 1988 1972/73 -
1984/85 12 no* 19-70 nurse-led
interview not reported
Krakow,
Poland Krzyzanowski,
90 16 1968 - 1981 1968 -
1981 13 yes† 19-70 interview not reported
Krakow,
Poland Krzyzanowski,
92 17 1968 - 1981 1968 -
1981 13 yes 19-70 interview not reported
Vlagtwedde/
Vlaardingen,
Holland Xu, 97 18 1965 - 1990 1965 -
1990 3 yes¶ 15-64 interview not reported
The Obstructive Lung Disease in Northern Sweden (OLIN)
Study Norrbotten,
Sweden Rønmark, 97 19 1985 - 1996 1985 -
1992 6 yes 35-36; 50-
51; 65-66 mailed
questionnaire - The Obstructive Lung Disease
in Northern Sweden (OLIN)
Study Norrbotten,
Sweden Lundback, 01 20 1985 - 1996 1985 -
1996 10 yes 35-36; 50-
51; 65-66 mailed
questionnaire - The Coronary Artery Risk
Development in Young Adults
(CARDIA) study Four centers
in the USA Beckett, 01 21 1985/86 -
1995/96 1985/86 -
1995/96 10 no 18-30 interview not reported
The Copenhagen City Heart
Study Copenhagen,
Denmark Godtfredson,
01 22 1976 - 1994 1976 -
1994 5/10 yes 20+ questionnaire -
The Tecumseh Health Study
Tecumseh, Michigan,
USA Broder, 74 23 1959 - 1965 1959 -
1965 4
not
applicable¥ all ages interview not reported Harjavalti,
Finland Huhti, 80 24 1961 - 1971 1961 -
1971 10 probably√ 40-64 nurse-led
interview not reported Lebanon,
Conneticut,
USA Beck, 82 25 1972 - 1978
1972 -
1978 6 no
"7-55+"
(?!) interview not reported
NHANES I USA McWhorter, 89 29 1971/5 -
1982/4 1971/5 -
1982/4 7-12 no 25-74 interview not reported
The European Community Respiratory Health Survey
(ECRHS) Three centers
in Sweden Plaschke, 00 28 1990 - 1993 1990 -
1993 3 yes 20-44 mailed
questionnaire - The European Community
Respiratory Health Survey (ECRHS)
Five centers
in Spain Basagna, 01 26
1991/93 - 1998/99
1991/93 -
1998/99 7 yesç 20-44 interview not reported
The Hordaland County Cohort study
Hordaland, Western
Norway Eagan, 02 27 1985 -
1996/97 1985 -
1996/97 11 yes 15-70 mailed
questionnaire -
The British National Child Development study
England, Scotland,
and Wales Strachan, 96 30 1958 - 1991 1975 -
1991 16 no
all subjects born march
3-9, 1958 interview not reported
The Finnish Twin Cohort StudyFinland Vesterinen, 88 31 1975 - 1990 1975 -
1981 6 no 18-64 mailed
questionnaire -
The Finnish Twin Cohort StudyFinland Houvinen, 99 32 1975 - 1990 1975 -
1990 15 no 18-45 mailed
questionnaire -
The Nurses Health Study The USA Troisi, 95 33 1976 - 1990 1980 -
1990 10 no 30-55 mailed
questionnaire -
Western
Sweden Toren, 99 34 1993 yes 20-50
mailed
questionnaire - The European Community
Respiratory Health Survey
(ECRHS) 16 countries Sunyer, 99 35 1991/93 yes 20-44 interview not reported
The Italian Study on Asthma
in Young Adults Nine Italian
centers deMarco, 02 36 1998/2000 yes 20-44 mailed
questionnaire - Hordaland,
Western
Norway Brogger, 04 37 1998 yes 15-70
mailed
questionnaire -
Register-based surveys
Rochester,
Minnesota, 1964 - not
¥
nurses examined medical
RETROSPECTIVE Cohort studies from a general population with more than one follow-upPROSPECTIVE Cohort studies from selected populations Cohort studies from a general population with only one follow-up Cross-sectional studies
Exact wording of questions reported
Similar wording of questions at baseline and follow-up
Size of baseline study population
Response rate, baseline
% of original study sample on which the incidence rates
are based Outcomes Predictors examined
yes probably** 2989 households approximately
55% *** not reported, but
less than 50% Asthma, attacks of dyspnea with wheeze
no probably** 2989 households approximately 55% ***
not reported, but less than 25%
(n=640)
'Asthma syndrome', chronic cough, chronic phlegm, chronic bronchitis, wheeze, attacks of dyspnea, dyspnea
no probably** 2989 households approximately 55% ***
not reported, but approximately
36% (n=1452) Asthma, chronic cough, chronic phlegm,
wheeze, attacks of dyspnea, dyspnea sex, age, smoking, city
no probably**
approximately 4633 (exact number not
reported) approximately
66% (n=3047) approximately 37.5% (n=1738)
'Asthma syndrome', chronic cough, chronic phlegm, chronic bronchitis, wheeze, attacks of dyspnea, dyspnea
no probably**
approximately 4633 (exact number not
reported) 94% (n=4355)
approximately 66.5%
(n=3082)
Asthma, chronic cough, chronic phlegm,
wheeze, attacks of dyspnea, dyspnea sex, age, smoking, city no probably** numbers diverge not reported approximately
31.1%
chronic cough, chronic phlegm, dyspnea gr.3, persistent wheeze, attacks of dyspnea with wheeze, bronchitis
BHR, adjusted for sex, age, smoking, and area of residence
yes yes 6610 86% 81.6% asthma, wheeze, attacks of dyspnea
sex, age, family history, population density, coastal/inland, smoking
yes yes 6610 86% 78.5% asthma
sex, age, family history, smoking, coastal/inland (and univariately SES)
no probably** 10143 51.0% 40.3% asthma sex, educational level,
smoking, BMI
yes probably** 19698 72.2% 51.8%/34.6% µasthma smoking
no no 9823 88.0% 66.8% asthma, allergic rhinitis allergic rhinitis
no probably** 1686 96.1% 69.0% chronic bronchitis, respiratory symptoms smoking
no probably** unknown∆ not reported not reported, but
less than 50% respiratory symptoms sex, smoking
yes yes not reported not reported unknown asthma, COPD sex, age, smoking,
income, race
yes yes 2422 76.8% 56.1% asthma
sex, smoking, pets at home, atopy, age, allergic rhinitis
yes yes 3370
approximately
58%ç 41.8% asthma
sex, age, IgE, BHR, symptoms, smoking, occupation
yes yes∫ 3786 89.0% 74.5% asthma, 10 symptoms sex, age, smoking
yes∑ no 18 559 78.5% 31.3% wheezing illness Many∑∑
yes yes
approximately 15 000 subjects (exact number
not reported) approximately
87% approximately
72% asthma sex, age, smoking
yes yes
approximately 15 000 subjects (exact number
not reported) approximately
87% approximately
58% asthma sex, age, hay fever,
chronic bronchitis no not reported 74 072 not reported impossible to
calculate asthma smoking
yes not applicable 20 000 79.1% 79.1% asthma
sex, age, smoking, family history, hay fever, atopic dermatitis
yes not applicable 30 000 63% 58.8% asthma sex, age of onset
yes not applicable 25 969 72.7% 72.7% asthma sex, age of onset
yes not applicable 2864 88.3% 88.3% asthma sex, age, smoking
yesß not applicable 57 890 subjects in 1980
unknown % of available records examined
unknown % of available records
examined asthma, single wheezing episodes sex, age
The perhaps most important quality measures next to random sampling, are the response rates. An extensive literature search was necessary to obtain the response rates (39-53), as there seem to be little agreement on how to report the response rates. For some of the studies, the method-sections of several papers describing each study were inadequate for exact calculation of the response rates. The most often under-reported response rates were the actual number of responders at baseline or the actual number of study subjects invited (15-18, 25, 31-33, 38). For the cohort studies, the incidence rates are based on subjects responding to one or more follow-ups. The most important measure of response is the percent of the original study sample on which the incidence rates were based. The numbers in table 1 reveal that many studies have rather low response rates.
The four cross-sectional studies and the register-based study were retrospective, and as such open to recall bias.
Ideally, a study should be prospective, of randomly sampled study subjects, with at least two time points, similar methodology at baseline and follow-up, and a high response rate.
Only two such studies have been conducted besides the Hordaland County Cohort Study;
namely the study from Krakow, Poland (17), and the Obstructive Lung disease In
** Not stated.
*** Only 1655 households out of the 2989 originally sampled.
† Only subjects with spirometry at 2 follow-ups were included.
¶ Of the two centers, one had an open cohort. How new cases were selected is not reported.
∑ All questions except the asthma question at last follow-up.
µ Godtfredson estimates incidence between baseline and first follow-up and between baseline and second follow-up.
¥ All inhabitants in geographical area were included in the cohort.
√ 'Non-selected'.
∆ (Due to lack of random sampling). 2272 subjects were included at baseline.
∫ For asthma and all symptoms except wheeze.
ß Exact set of criteria is given.
* Random cluster of households, but stratified by age, ethnicity, and socio-economic status. Weights seems not to have been used when calculating incidence rates.
ç Baseline is actually a selection of responders to an earlier survey, thus 16% were selected away before baseline, in a non-random manner.
∑∑ Sex, several factors pertaining to maternal health in pregnancy, complications during birth, breast feeding, longitude and latitude, degree of urbanisation, pollution, other diseases, pneumonia, whooping cough, tonsillectomy, hay fever, eczema, abdominal pain, vomitting, and migraine.
Northern Sweden (OLIN) Study (19, 20). Only the study from Krakow, Poland had been published at the time the Hordaland County Cohort Study follow-up was planned.
Remission
A few of the studies in table 1 have also examined the remission of asthma or respiratory symptoms (17, 18, 28, 36, 54, 55). In the current paradigm of asthma as a chronic
disease, it is a matter of debate whether ‘true’ remission of the airways inflammation really occurs. Regardless, some subjects with a label of asthma at one time will state at a later time that they no longer have asthma. It is of interest to know the size of that
fraction, and perhaps more so, which factors influence a migration out of the diagnosis of asthma. Estimating this fraction (which can be defined as the cumulative remission) is fraught with some serious questions regarding methodology, which are discussed later.
The Hordaland County Cohort Study did not allow for a reasonable estimation of the remission of asthma, due to the wording of the asthma question at follow-up. A
reasonable estimation of the remission of respiratory symptoms was possible; previously only conducted in the Tucson and Krakow cohorts (17), and later in the
Vlagtwedde/Vlaardingen cohort (18).
Risk factors
Defining asthma is difficult. The clinical picture varies from patient to patient, and over time, in each patient. There is no ‘gold-standard’ test. Not surprisingly then, are the causes of asthma still not fully understood. Whether asthma is one or several diseases can be debated. Before, asthma was often thought of as ‘allergic’, or ‘non-allergic’. The current consensus defines asthma as a chronic inflammatory disorder of the airways, associated with airway hyper-responsiveness, airflow limitation, and respiratory symptoms (56). Asthma is likely to be a multi-factorial disease, with several pre- disposing host factors and environmental factors.
A cohort study has the advantage of enabling an assessment of a risk factor before the incident event; a requirement if the risk factor is to be thought of as causal. If the effect of the risk factor is not strong, a larger sample is often needed to demonstrate the
association. A selected cohort, of for instance specific occupational groups, can be of
great value in evaluation of potential risk factors. However, with a selected cohort there is the question whether the findings are applicable to the population at large. If the
preventive potential of the risk factor is to be assessed in the whole population, a general population study is the preferred design.
Thus, the Hordaland County Cohort Study enabled the assessment of some risk factors for the incidence of respiratory symptoms and asthma, as well as their preventive potential.
With recent advances in computational technology, sophisticated analyses of the impact of several risk factors have become available. Whereas studies from the 1970’s rarely had adjusted for more than one co-factor, recent studies often take several risk factors into account, usually by some sort of regression analyses. In table 1, the potential risk factors examined in each study are listed. The factors listed are either risk factors for which the studies have determined the potential effect of, or as a confounder to be adjusted for.
Smoking is the most commonly assessed risk factor. Where several study centers are involved, adjustment for study center is usually performed. Family history and/or atopy are often taken into account, atopy being the strongest known risk factor for asthma.
Some assessment or adjustment for socio-economic status was performed in two of the studies listed in table 1 (20, 21). Only one study examined the effect of occupational exposure (26). One other paper from the cohort from Krakow, Poland, specifically examined the effect of occupational exposure on the incidence of asthma and respiratory symptoms (57).
Krzyzanowski et al examined the remission of respiratory symptoms by sex, age, and smoking habits (17). Xu et al examined the remission of respiratory symptoms by bronchial hyper-responsiveness (BHR) (18).
AIMS
The aims of the Hordaland County Cohort Study were:
1) To examine predictors for response at follow-up, and the consequences of increasing the response rates through several reminders.
2) To measure the occurrence of new cases of asthma and respiratory symptoms by sex and age in the adult general population in Norway.
3) To describe the effect of the risk factors smoking, occupational exposure, and educational level, on the incidence of respiratory symptoms and asthma.
4) To examine the remission of respiratory symptoms, and the effect of changes in smoking habits, previous occupational exposure, and educational level.
MATERIALS AND METHODS
Study design
The baseline survey was conducted in the fall of 1985, and has been well described previously (4, 58). All adults aged 15-70 years, in the county of Hordaland, were eligible for inclusion in the survey, comprising a total number of 267 403 individuals per 31st of December 1984. All Norwegian citizens are given a unique identification number at birth, and county registers have list of addresses for all citizens, based on these numbers. A random sample of 4992 subjects (or 1.87%) was drawn by the Norwegian Central Bureau of Statistics (4).
The follow-up survey was conducted between September of 1996 and May 1997. The survey was expanded, to allow for a voluntary examination at the Department of Thoracic Medicine, Haukeland University Hospital2, as well as several questionnaires distributed only among those subjects who attended this examination. Due to the distances involved in traveling within Hordaland County, only subjects living in the city of Bergen and 11 surrounding municipalities were eligible for follow-up (table 2).
2 The current dissertation is based on information sampled by mailed questionnaires only.
Municipality Women Men Total
Bergen 73001 71480 144481
Voss 4653 4768 9421
Samnanger 728 777 1505
Os 3734 3845 7579
Sund 1381 1447 2828
Fjell 3829 4056 7885
Askøy 5809 6068 11877
Vaksdal 1484 1593 3077
Osterøy 2064 2286 4350
Øygarden 820 961 1781
Radøy 1582 1682 3264
Lindås 3329 3575 6904
Sum 102414 102538 204952
Table 2. Number of inhabitants in Bergen and 11 surrounding municipalities, aged 15-70, per 31st of December 1984.
Of the initial 4992 subjects, 3786 were living in Bergen or one of the 11 surrounding municipalities per 31st of December 1984. Of these 3786 subjects, 3370 had responded to the baseline survey. Between 1985 and September 1996, 189 subjects were deceased, leaving 3181 subjects eligible for the follow-up survey (Figure 1).
Figure 1. Flowchart of the Hordaland County Cohort Study
Data collection
In 1985, all 4992 subjects received by mail a one-sheet questionnaire (Appendix A), together with a letter explaining the survey, and a paid reply envelope. If no reply had been obtained within three weeks, a new mailing (with a new questionnaire, letter, and
paid reply envelope) was performed. This was repeated one final time, thus each subject received the original letter and up to two reminders.
The questionnaire was expanded from 40 to 58 questions at follow-up (Appendix B), making room for questions regarding health care utility, environmental tobacco smoke, indoor climate, and family history. Altogether, the 1996/97-questionnaire comprised four pages.
Starting in September 1996, the subjects received by mail a letter explaining the intention of the follow-up phase of the survey, the questionnaire, a paid reply-envelope, and an invitation to the examination, scheduled approximately two weeks thereafter. If an individual did not show up at the examination, they were rescheduled, and sent a new letter with the new time. If this individual also had not returned the questionnaire by mail, a new questionnaire (and paid reply-envelope) was included in the reminder letter. This was repeated for a second reminder if necessary. If the individual had not responded after two reminder letters, one of three physicians from the Department of Thoracic Medicine would try to reach the individual by phone. Thus, there were the following categories of responders: Those responding only by mail, those attending after receiving the initial questionnaire, those attending after receiving the first reminder, those attending after receiving the second reminder, those attending after a telephone reminder, and finally those not responding at all. Data collection was finalized in May 1997.
Of the 3181 study subjects, 2819 (88.6%) returned the questionnaire (figure 1) after all reminders.
Processing the data file
Inconsistencies
The processing of the baseline data is previously described (4). The follow-up data was initially typed into Epi Info3, and contained some programming to prevent wrongful
3 Epi Info is a free program developed by the Centers for Disease Control (CDC), and can be obtained at: http://www.cdc.gov/epiinfo
typing. After all the data had been typed and transferred to SPSS, we conducted analyses of inconsistencies. For instance, if a subject had answered ‘no’ to ever having been treated by a physician, or at a hospital for asthma (question 21), but had given a specific year for age of debut of the disease, it would be considered an inconsistency. (This happened to have occurred in one instance for this particular question).
Based on the questionnaire, we constructed all theoretical inconsistencies we could think of, a total number of 50. Most of these were actually inconsistent missing values, for example saying ‘yes’ to having been treated for asthma, but leaving the space for ‘age at debut’ open. We then identified all individual inconsistencies, and manually checked all affected questionnaires. By then, either a true inconsistency existed in the questionnaire, or a typing error was present.
Previous to the identification of the affected questionnaires, we constructed rules for interpretation in the case of an inconsistency. For example, in the case on question 21 where a subject had stated no to being treated for asthma, but had given an age for debut of the disease, we interpreted the age as more specific, and the subject’s answer to question 21 would be labeled ‘interpreted as yes’. Obviously, if the inconsistency was merely a missing value, no changes were made.
Not considering inconsistent missing values, there were a total of 30718 potential individual inconsistencies. Altogether, 79 individual inconsistencies were found, or 0.26%. All the inconsistency-analyses conducted, lists of affected questionnaires, and rules for interpretation have been recorded, and stored, both digitally and on paper.
Typing errors
Although Epi Info was programmed to not allow wrongful typing of for instance a number not predefined, it was still possible to type ‘2’ (for no) instead of ‘1’ (for yes).
After completion of the analyses on possible inconsistencies, we conducted a small-scale evaluation of typing errors. All questionnaires from each subject were identifiable by a unique number. The data was entered according to their numbering, so that the
questionnaire number 1 was the first to be typed, and number 3370 the last. Fifty questionnaires, from five different time-periods of typing (#181-190, 861-870, 1541-
1550, 2221-2230, 2901-2910), were chosen for closer examination of possible typing errors. No systematic errors were found, in that no particular question seemed prone for errors. Out of 3280 possible errors overall, only 31 were found. For the first 19 questions (regarding respiratory symptoms), no errors were found at all.
Missing values
Overall, the number of missing values was low. For the outcome variables, the number of missing values varied between 0.3% (morning cough – follow-up survey) to 2.4%
(dyspnea grade 1 – baseline survey). The three main exposures examined in this dissertation are smoking, occupational exposures and education. Whereas almost all subjects reported whether they were smoking daily (0.9% missing values at baseline and 0.2% at follow-up), more subjects did not report numbers of year smoked (4.9% at baseline and 2.8% at follow-up) or numbers of cigarettes per day (5.6% at baseline and 4.7% at follow-up). 1.0% of responders at baseline did not answer the question about previous dust or fumes exposure, and 2.2% of the responders at follow-up failed to answer the question about educational level.
In the baseline survey, missing values were significantly more prevalent among women than men for the outcome variables, whereas no differences were found at follow-up, or for the exposure variables at either baseline or follow-up. For most of the questions, missing values were more prevalent with increasing age.
Variables
The outcome variables were categorized as dichotomous variables. ‘Yes’ and ‘interpreted as yes’ (if existent) were grouped into yes, and ‘no’ and ‘interpreted as no’ into no.
Missing values were interpreted as no. For the exposure variables, it was felt imprudent to interpret the missing values, and they were left as missing. Thus, for some analyses, subjects with missing values would be excluded.
Outcome variables were questions 4 through 14, and question 26, in the baseline survey (appendix A), and question 1 through 13, and question 21, at follow-up (appendix B).
In question number 15 and 16 in the baseline survey, and number 34 and 36 in the follow-up survey, the subjects stated whether they were daily smokers, or former
smokers. Based on these questions, a three-category variable was created, with categories never-, ex-, current- smokers.
One pack-year was defined as the equivalent of smoking 20 cigarettes per day for one whole year. This was calculated by combining question 17 and 18 in the baseline survey.
In 1985, smoking only pipe tobacco was still not completely uncommon (among men). In subjects smoking pipe tobacco, one pack year was defined as smoking 50 gram (one pack) of tobacco per week. This information was available by combining question 17 and 20. By the time of the follow-up survey, pipe tobacco was considered to be as rare as to be irrelevant to the analyses, and the questions on pipe tobacco left out of the
questionnaire.
The analyses on occupational exposures are based on question number 34, 35, and 39 in the baseline survey.
The analyses on educational level are based on question number 56 in the follow-up survey.
Statistical analyses
The main outcome was incident cases of respiratory symptoms or asthma. The two surveys were eleven years apart. At which time the incident event occurred within this eleven year time period, cannot be known from these two questionnaires. Incidence was defined as the number of new cases occurring in the study population during the follow- up time period divided by the number of persons at risk, i.e. subjects who did not have the symptom in question at baseline. Thus, the main outcome was the cumulative incidence over 11 years.
The cumulative remission of any symptom was defined as the proportion of subjects having a symptom at baseline who no longer reported the symptom at follow-up.
The main measure of association was the odds ratio (59). For evaluation of adjustment, all odds ratios were calculated by logistic regression. In the first three papers, SPSS (7.5- 10.0) was used for computation (60, 61).
An important measure of effect is the attributable fraction (62). It has been a problem to compute confidence intervals for adjusted attributable fractions. However, with a module in Stata 6.0 (63) we were able to compute the confidence intervals, which are presented in paper III (64). Subsequent computation was mainly conducted in Stata 7.0–8.0 (65).4
4 After this researcher had been exposed to Stata, it quickly became the statistical program of choice.
SYNOPSIS OF PAPERS
Paper IThe overall worry with non-response is that those subjects not participating have a different chance of disease than those participating. In several cross-sectional studies, analyses of non-responders have often showed that they tend to be slightly different than responders with regard to some important variables, like socioeconomic status, smoking, and disease.
Thus, we asked whether demographic characteristics, or respiratory symptoms or asthma at baseline, would influence the response rates at the follow-up phase. Further, we examined whether the observed exposure-disease relationships were altered if the analyses were conducted with only the early responders, only the late responders, or all responders.
Among the non-responders in 1996/97 (follow-up), 12.7% were unemployed in 1985 (baseline) compared to 7.2% of the responders. 9.8% of the non-responders were retired in 1985, but only 3.3% of the responders. Whereas 72.6% of the responders in 1996/97 had been early responders in 1985, only 52.5% of the non-responders in 1996/97 had been early responders in 1985.
These relationships held up in the multivariate analyses. The OR (95% CI) for response at follow-up was 0.38 (0.25, 0.56) among the subjects who needed two reminders in 1985, compared to the early responders in 1985. Furthermore, being unemployed, retired, or a student in 1985 was predictive of a lower chance for response in 1996/97. Sex, age, smoking status in 1985, and respiratory health in 1985 did not influence the risk for non- response at follow-up.
The effect of smoking on the incidence of respiratory symptoms was slightly different when the analyses were conducted on only the late responders, compared to only the early responders. However, the overall impact was small, as the ORs for the incidences were nearly identical when comparing only the early responders, to all responders (table
4, paper I). For gender and age, there was no discernable difference in the ORs obtained by inclusion of various groups of responders.
Paper II
A large amount of papers have been published in the later years, describing a rise in the prevalence of asthma. Less is known about the true incidence of asthma, as there are fewer longitudinal studies from general populations. Asthma is difficult to define accurately, both clinically and in epidemiological studies, and the diagnosis is in large part dependent on symptomatology.
The objective of the second paper was to assess incidence rates for a wide range of respiratory symptoms and asthma, both overall and within sex-, age-, and smoking groups.
In 1985, the prevalence of asthma was 3.4% in the study sample. This increased to 6.0%
at follow-up, however with a cohort now 11 years older. The cumulative incidence of asthma within the 11-year follow-up was 3.7%. The cumulative incidence of the cough symptoms varied between 9.0% for chronic cough to 16.5% for phlegm cough. The cumulative incidence of dyspnea grade 2 was 11.9%, and for attacks of dyspnea 10.3%.
Women had a higher risk for developing morning cough, chronic cough, any degree of dyspnea, attacks of dyspnea, and wheezing, but not asthma. The risk for developing wheezing and attacks of dyspnea decreased with age, whereas the chance of all other symptoms increased with age. Those subjects who started to smoke within the follow-up period (non-current), or smoked throughout (current-current), had an increased risk for developing wheezing and all cough symptoms, but not asthma or degrees of dyspnea.
However, the subjects with the greatest smoking load, measured in pack-years, had a significantly increased risk for the incidence of both asthma and dyspnea, as well as wheezing and cough.
Paper III
In several industries, it has been shown that occupational exposures are a risk factor for obstructive lung disease. The impact of occupational exposure on the disease burden in the population will vary according to the composition of different industries in the specific population, as well as working regulations, controlling individual safety measures and the like.
The aim of the third paper was to assess the burden of asthma and respiratory symptoms attributable to occupational exposure in the general population of Hordaland County.
In 1985, 43.5% of the men reported ever having had a workplace with exposure to dust or fumes, and 13.0% of the women. The corresponding figures for quartz and asbestos exposures were 7.3% and 10.1% for men, and only 0.4% and 0.4% for women.
Among the subjects who did not report any previous occupational exposures in 1985, 3.2% later developed asthma, compared to 5.3% among subjects who did report previous dust or fumes exposure. Among subjects with previous asbestos exposure, 7.5% later developed asthma.
The same trend was seen for all respiratory symptoms examined; chronic cough, phlegm cough, dyspnea grade 3, attacks of dyspnea, and wheezing. The cumulative incidence was increased in subjects reporting dust or fumes exposures prior to baseline.
After adjustment for sex, age, educational level, smoking habits (never, ex, current), and pack-years among those having smoked, the risk of developing respiratory symptoms varied between an OR (95% CI) of 1.4 (1.1, 1.7) for wheezing, to 2.1 (1.3, 3.2) for dyspnea grade 3 for those previously exposed to dust or fumes, compared to those not exposed. The OR (95% CI) for developing asthma among those exposed was 1.6 (1.01, 2.5).
The adjusted attributable fraction (95% CI) of occupational exposure on the incidence of asthma was 14.4% (-1.2, 27.6). For the respiratory symptoms, the adjusted attributable fractions varied between 5.7% (1.1, 10.0) for wheezing to 19.3% (6.5, 30.4) for dyspnea grade 3.
Paper IV
Studies on risk factors for the incidence of a disease or symptom, is also a study of preventive potential. Studies on risk factors for remission are studies on the potential remedy, once a disease or symptom has occurred. Although the beneficial effect of smoking cessation on lung function decline has been shown in several longitudinal studies, only one previous cohort study had specifically studied the effect of smoking cessation to the remission of respiratory symptoms. No previous cohort study had examined the effect of previous occupational exposure to the remission of respiratory symptoms.
Thus, the aim of the fourth paper was to assess the cumulative remission of selected respiratory symptoms, and the factors that influenced the chance for remission.
The measured cumulative remission varied from 42.3% for morning cough to 58.4% for chronic cough. Younger subjects had higher remission rates for all symptoms, except wheezing. The cumulative remission of morning cough was 72.2% among those subjects giving up smoking in the follow-up period, and only 33.9% in those not giving up smoking. Among those ever having smoked, the lowest remission was seen among those having smoked more than 20 pack-years. For all symptoms, subjects with previous occupational exposure had a lower cumulative remission than subjects without occupational exposure.
The OR (95% CI) for remission of dyspnea grade 2 was 2.0 (1.01, 3.8) in men compared to women, after adjustment for age, educational level, smoking, and dust or fumes exposure. With increasing age, there was a lower chance of remission of the cough symptoms and dyspnea grade 2, after adjustment.
The OR (95% CI) for remission of the cough symptoms varied from 2.6 (1.5, 4.5) for phlegm cough to 6.2 (3.5, 11.2) for morning cough, in subjects who gave up smoking, compared to persisting smokers. Smoking cessation was beneficial also for the remission of wheezing, with an OR (95% CI) of 2.2 (1.3, 3.7), whereas for the remission of dyspnea grade 2 and attacks of dyspnea, smoking cessation had no significant effect.
Not having previous dust or fumes exposure was associated with an increased chance of remission for all symptoms, however not statistically significant for chronic cough and attacks of dyspnea.
Paper V
It is well known that lower socioeconomic status is associated with poorer health.
Socioeconomic status is a term not easily defined, but is usually taken to describe ones social standing, by terms of income, education, and job opportunities. Conflicting results have been found when examining the effect of socioeconomic status on asthma in children. For adults, very few studies exist examining the relationship between incident asthma and socioeconomic status.
Using educational level as a marker of socioeconomic status, and adjusting for occupational exposures, as well as for sex, age, atopy, and smoking, we wanted to examine the effect of educational level on the incidence of respiratory symptoms and asthma.
Educational level was categorized as primary, secondary, or university, based on type and length of schooling. In 1996/97, the cohort consisted of subjects aged 26-81 years. By then, 18.2% had a primary, 55.7% a secondary, and 26.1% a university educational level.
More men (28.6%) than women (23.8%) had a university level education, as well as more never-smokers (33.4%) had a university level education compared to current smokers (18.2%).
The cumulative incidences of all symptoms examined were higher among subjects with the lowest educational level. Whereas the cumulative incidences ranged from 5.7%
(dyspnea grade 2) to 13.6% (wheezing) among subjects with a university level education, they were generally higher among subjects with a primary educational level, ranging from 11.5% (chronic cough) to 20.4% (morning cough). The cumulative incidence of asthma was 1.8%, 4.1%, and 5.3% among subjects with a university-, secondary-, and primary educational level, respectively.
Whether one had a primary or secondary educational level did not matter as much as the difference observed for those with a university level education. The OR for the incidence of all symptoms and asthma was higher in those with a primary educational level,
compared to those with a university educational level, after adjustment for sex, age, atopy, smoking, and occupational exposure.